1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2020 Facebook */
3 #include "bpf_iter.h"
4 #include "bpf_tracing_net.h"
5 #include <bpf/bpf_helpers.h>
6 #include <bpf/bpf_endian.h>
7
8 char _license[] SEC("license") = "GPL";
9
hlist_unhashed_lockless(const struct hlist_node * h)10 static int hlist_unhashed_lockless(const struct hlist_node *h)
11 {
12 return !(h->pprev);
13 }
14
timer_pending(const struct timer_list * timer)15 static int timer_pending(const struct timer_list * timer)
16 {
17 return !hlist_unhashed_lockless(&timer->entry);
18 }
19
20 extern unsigned CONFIG_HZ __kconfig;
21
22 #define USER_HZ 100
23 #define NSEC_PER_SEC 1000000000ULL
jiffies_to_clock_t(unsigned long x)24 static clock_t jiffies_to_clock_t(unsigned long x)
25 {
26 /* The implementation here tailored to a particular
27 * setting of USER_HZ.
28 */
29 u64 tick_nsec = (NSEC_PER_SEC + CONFIG_HZ/2) / CONFIG_HZ;
30 u64 user_hz_nsec = NSEC_PER_SEC / USER_HZ;
31
32 if ((tick_nsec % user_hz_nsec) == 0) {
33 if (CONFIG_HZ < USER_HZ)
34 return x * (USER_HZ / CONFIG_HZ);
35 else
36 return x / (CONFIG_HZ / USER_HZ);
37 }
38 return x * tick_nsec/user_hz_nsec;
39 }
40
jiffies_delta_to_clock_t(long delta)41 static clock_t jiffies_delta_to_clock_t(long delta)
42 {
43 if (delta <= 0)
44 return 0;
45
46 return jiffies_to_clock_t(delta);
47 }
48
sock_i_ino(const struct sock * sk)49 static long sock_i_ino(const struct sock *sk)
50 {
51 const struct socket *sk_socket = sk->sk_socket;
52 const struct inode *inode;
53 unsigned long ino;
54
55 if (!sk_socket)
56 return 0;
57
58 inode = &container_of(sk_socket, struct socket_alloc, socket)->vfs_inode;
59 bpf_probe_read_kernel(&ino, sizeof(ino), &inode->i_ino);
60 return ino;
61 }
62
63 static bool
inet_csk_in_pingpong_mode(const struct inet_connection_sock * icsk)64 inet_csk_in_pingpong_mode(const struct inet_connection_sock *icsk)
65 {
66 return icsk->icsk_ack.pingpong >= TCP_PINGPONG_THRESH;
67 }
68
tcp_in_initial_slowstart(const struct tcp_sock * tcp)69 static bool tcp_in_initial_slowstart(const struct tcp_sock *tcp)
70 {
71 return tcp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
72 }
73
dump_tcp6_sock(struct seq_file * seq,struct tcp6_sock * tp,uid_t uid,__u32 seq_num)74 static int dump_tcp6_sock(struct seq_file *seq, struct tcp6_sock *tp,
75 uid_t uid, __u32 seq_num)
76 {
77 const struct inet_connection_sock *icsk;
78 const struct fastopen_queue *fastopenq;
79 const struct in6_addr *dest, *src;
80 const struct inet_sock *inet;
81 unsigned long timer_expires;
82 const struct sock *sp;
83 __u16 destp, srcp;
84 int timer_active;
85 int rx_queue;
86 int state;
87
88 icsk = &tp->tcp.inet_conn;
89 inet = &icsk->icsk_inet;
90 sp = &inet->sk;
91 fastopenq = &icsk->icsk_accept_queue.fastopenq;
92
93 dest = &sp->sk_v6_daddr;
94 src = &sp->sk_v6_rcv_saddr;
95 destp = bpf_ntohs(inet->inet_dport);
96 srcp = bpf_ntohs(inet->inet_sport);
97
98 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
99 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
100 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
101 timer_active = 1;
102 timer_expires = icsk->icsk_timeout;
103 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
104 timer_active = 4;
105 timer_expires = icsk->icsk_timeout;
106 } else if (timer_pending(&sp->sk_timer)) {
107 timer_active = 2;
108 timer_expires = sp->sk_timer.expires;
109 } else {
110 timer_active = 0;
111 timer_expires = bpf_jiffies64();
112 }
113
114 state = sp->sk_state;
115 if (state == TCP_LISTEN) {
116 rx_queue = sp->sk_ack_backlog;
117 } else {
118 rx_queue = tp->tcp.rcv_nxt - tp->tcp.copied_seq;
119 if (rx_queue < 0)
120 rx_queue = 0;
121 }
122
123 BPF_SEQ_PRINTF(seq, "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X ",
124 seq_num,
125 src->s6_addr32[0], src->s6_addr32[1],
126 src->s6_addr32[2], src->s6_addr32[3], srcp,
127 dest->s6_addr32[0], dest->s6_addr32[1],
128 dest->s6_addr32[2], dest->s6_addr32[3], destp);
129 BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d ",
130 state,
131 tp->tcp.write_seq - tp->tcp.snd_una, rx_queue,
132 timer_active,
133 jiffies_delta_to_clock_t(timer_expires - bpf_jiffies64()),
134 icsk->icsk_retransmits, uid,
135 icsk->icsk_probes_out,
136 sock_i_ino(sp),
137 sp->sk_refcnt.refs.counter);
138 BPF_SEQ_PRINTF(seq, "%pK %lu %lu %u %u %d\n",
139 tp,
140 jiffies_to_clock_t(icsk->icsk_rto),
141 jiffies_to_clock_t(icsk->icsk_ack.ato),
142 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(icsk),
143 tp->tcp.snd_cwnd,
144 state == TCP_LISTEN ? fastopenq->max_qlen
145 : (tcp_in_initial_slowstart(&tp->tcp) ? -1
146 : tp->tcp.snd_ssthresh)
147 );
148
149 return 0;
150 }
151
dump_tw_sock(struct seq_file * seq,struct tcp_timewait_sock * ttw,uid_t uid,__u32 seq_num)152 static int dump_tw_sock(struct seq_file *seq, struct tcp_timewait_sock *ttw,
153 uid_t uid, __u32 seq_num)
154 {
155 struct inet_timewait_sock *tw = &ttw->tw_sk;
156 const struct in6_addr *dest, *src;
157 __u16 destp, srcp;
158 long delta;
159
160 delta = tw->tw_timer.expires - bpf_jiffies64();
161 dest = &tw->tw_v6_daddr;
162 src = &tw->tw_v6_rcv_saddr;
163 destp = bpf_ntohs(tw->tw_dport);
164 srcp = bpf_ntohs(tw->tw_sport);
165
166 BPF_SEQ_PRINTF(seq, "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X ",
167 seq_num,
168 src->s6_addr32[0], src->s6_addr32[1],
169 src->s6_addr32[2], src->s6_addr32[3], srcp,
170 dest->s6_addr32[0], dest->s6_addr32[1],
171 dest->s6_addr32[2], dest->s6_addr32[3], destp);
172
173 BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
174 tw->tw_substate, 0, 0,
175 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
176 tw->tw_refcnt.refs.counter, tw);
177
178 return 0;
179 }
180
dump_req_sock(struct seq_file * seq,struct tcp_request_sock * treq,uid_t uid,__u32 seq_num)181 static int dump_req_sock(struct seq_file *seq, struct tcp_request_sock *treq,
182 uid_t uid, __u32 seq_num)
183 {
184 struct inet_request_sock *irsk = &treq->req;
185 struct request_sock *req = &irsk->req;
186 struct in6_addr *src, *dest;
187 long ttd;
188
189 ttd = req->rsk_timer.expires - bpf_jiffies64();
190 src = &irsk->ir_v6_loc_addr;
191 dest = &irsk->ir_v6_rmt_addr;
192
193 if (ttd < 0)
194 ttd = 0;
195
196 BPF_SEQ_PRINTF(seq, "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X ",
197 seq_num,
198 src->s6_addr32[0], src->s6_addr32[1],
199 src->s6_addr32[2], src->s6_addr32[3],
200 irsk->ir_num,
201 dest->s6_addr32[0], dest->s6_addr32[1],
202 dest->s6_addr32[2], dest->s6_addr32[3],
203 bpf_ntohs(irsk->ir_rmt_port));
204 BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
205 TCP_SYN_RECV, 0, 0, 1, jiffies_to_clock_t(ttd),
206 req->num_timeout, uid, 0, 0, 0, req);
207
208 return 0;
209 }
210
211 SEC("iter/tcp")
dump_tcp6(struct bpf_iter__tcp * ctx)212 int dump_tcp6(struct bpf_iter__tcp *ctx)
213 {
214 struct sock_common *sk_common = ctx->sk_common;
215 struct seq_file *seq = ctx->meta->seq;
216 struct tcp_timewait_sock *tw;
217 struct tcp_request_sock *req;
218 struct tcp6_sock *tp;
219 uid_t uid = ctx->uid;
220 __u32 seq_num;
221
222 if (sk_common == (void *)0)
223 return 0;
224
225 seq_num = ctx->meta->seq_num;
226 if (seq_num == 0)
227 BPF_SEQ_PRINTF(seq, " sl "
228 "local_address "
229 "remote_address "
230 "st tx_queue rx_queue tr tm->when retrnsmt"
231 " uid timeout inode\n");
232
233 if (sk_common->skc_family != AF_INET6)
234 return 0;
235
236 tp = bpf_skc_to_tcp6_sock(sk_common);
237 if (tp)
238 return dump_tcp6_sock(seq, tp, uid, seq_num);
239
240 tw = bpf_skc_to_tcp_timewait_sock(sk_common);
241 if (tw)
242 return dump_tw_sock(seq, tw, uid, seq_num);
243
244 req = bpf_skc_to_tcp_request_sock(sk_common);
245 if (req)
246 return dump_req_sock(seq, req, uid, seq_num);
247
248 return 0;
249 }
250